Week 2025-36
product launch
- Another Giant Leap: The Rubin CPX Specialized Accelerator & Rack — SemiAnalysis, 2025-09-10Nvidia launched the Rubin CPX specialized accelerator designed for prefill inference tasks, noting its high compute FLOPS capability and comparing it to the GB200 NVL72 Oberon rack-scale form.
supply constraint
- Huawei Ascend Production Ramp: Die Banks, TSMC Continued Production, HBM is The Bottleneck — SemiAnalysis, 2025-09-08The production of AI compute is highly dependent on the supply chain, with high bandwidth memory (HBM) currently cited as the main bottleneck.